General structure of the Multi-layer Perceptron. The input layer consists of 1 passive node, which is the output of the preprocessing pipeline, which relays the values directly to the hidden layer. The hidden layers H¹ consist of M nonlinear active neurons that modify their input values and produce a signal to be fed into the next layer. The output layer then transforms its input values (from the hidden layer) into the flow values at target height z_t.